3_31_2019_detecting .pdf

Air & Waste

ISSN: 1073-161X (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/uawm19

Detecting and Tracking Changes in Ozone Air Quality S. Trivikrama Rao & Igor G. Zurbenko To cite this article: S. Trivikrama Rao & Igor G. Zurbenko (1994) Detecting and Tracking Changes in Ozone Air Quality, Air & Waste, 44:9, 1089-1092, DOI: 10.1080/10473289.1994.10467303 To link to this article: https://doi.org/10.1080/10473289.1994.10467303

Published online: 05 Mar 2012.

Submit your article to this journal

Article views: 607

Citing articles: 72 View citing articles

Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=uawm20

TECHNICAL PAPER ISSN 1047-3289/ Air & Waste Manage. Assoc. 44s 1089-1092

Detecting and Tracking Changes in Ozone Air Quality S. Trivikrama Rao and Igor G. Zurbenko State University of New York at Albany Albany, New York This paper presents a statistical method for filtering out or moderating the influence of meteorological fluctuations on ozone concentrations. Use of this technique in examining trends in ambient ozone air quality is demonstrated with ozone data from a monitoring location in New Jersey. The results indicate that this method can detect changes in ozone air quality due to changes in emissions in the presence of meteorological fluctuations. This method can be useful in examining the effectiveness of regulatory initiatives in improving ozone air quality.

Introduction It is well known that meteorology plays a significant role in establishing conditions conducive to the formation and buildup of high ozone concentrations.13 Analysis of ozone air quality trends in the northeastern United States is hampered by the lack of a longterm quality-assured database, as well as by the variability in meteorology. Also, it is difficult to detect the change in ozone air quality due to the change in emissions in the presence of meteorological fluctuations.4 Furthermore, unless the change in emissions is substantial, any improvement in ozone air quality achieved from modest emissions reductions can be easily masked by the variability in meteorology.5 Since ambient ozone concentrations are strongly influenced by meteorological fluctuations, statistically robust methods are needed to track the effectiveness of regulatory initiatives in improving ambient ozone air quality. We must separate different phenomena present in the time series of both meteorological and ozone data which have different characteristics such as long- and short-term variations. A precise examination of either meteorological or ozone data becomes extremely complex because both strongly affect each other and produce false effects for each side. Statistical methods currently being used68 to detect ozone trends are not powerful enough to clearly separate random variations and meteorological effects from long-term trends. In this paper, we present a method to filter out or moderate the influence of meteorology on ambient ozone levels, using surface temperature as a surrogate for all meteorological conditions that affect ozone. The results from the application of this method to ozone data from Cliffside Park, New Jersey, reveal a marked

Because the influence of stochastic and seasonal variations in meteoro

change in ambient ozone concentrations during the post-1988 period. This change in ozone concentrations in 1989 may be a result of controls on the volatility of fuel used in the Northeast. Data and Analysis Procedures Data Hourly concentrations of ozone measured at the Cliffside Park monitor in northern New Jersey from 1983 through 1991 were obtained from the U.S. EPA's Aerometric Information Retrieval System (AIRS) database. Also, hourly values of surface temperature measured at LaGuardia Airport in New York City, which is the weather station nearest to the location of the ozone monitor, were assembled for the above time period. A subset of data consisting of daily maxima of temperature and ozone was derived from the hourly time series in the following analysis. Method of Analysis We assume that a time series of temperature or ozone may be represented as X(t) = e(t) + S(t) + W(t) (1) where X(t) is the original time series, e(t) is a trend component, S (t) is seasonal variation, and W(t) is white noise. In our analysis, we will separate the deterministic portions (e and S) from the short-term variations (white noise) in the data using the Kolmogorov-Zurbenko (KZmp) filter.* The KZmp filter is a lowpass filter produced by repeated iterations of a simple moving average. The moving average (each iteration) is defined by

i

(2)

j=-K

assess the effectiveness of regulatory initiatives in improving quality. Although i meteorologically weighted measure of alleviate some of thedifficulties associated with the analysis of ozone trends, a statistically robust method Is necessary to Meet changes in ozone air Hogicat fluctuations.

where m = 2k + 1. The Yj become the input for the second pass, and so on. Determination of the final low-pass filter (specifying "m" and the number of passes "p") is an iterative process in which the data user determines that the white noise has been removed. The output time series, Yt, is the low-frequency part of Xt, Yt=KZm>p(X£).

(3)

Copyright 1994 - Air & Waste Management Association

Journal of Air & Waste Management Association • Vol. 44 • September 1994 • 1089

TECHNICAL PAPER The Yt contain both long-term trend and seasonal effects. The KZ filter and its statistical characteristics have been carefully studied by Zurbenko.10 When the KZ29,3 [29-day length (m) with three iterations (p)] filter is applied to the daily maximum temperature and the log of daily maximum ozone concentration data, one obtains time series that exhibit no white noise (Figure 1). The filtered temperature and the log of ozone time series are denoted hereafter as T^t) and 0^(1), respectively. Quantile-quantile (QQ) plots of the residual of the filtered temperature [T(t) - T^t)] and the log of ozone [{O(t) - Okz(t)} = W(t)] are presented in Figures 2(a) and 2(b), respectively, to illustrate that they are indeed white noise; a QQ plot of 3,000 normal random numbers is shown in Figure 2(c) for comparison. Periodograms (not shown here) of the residuals also confirm that they are almost white noise. In addition, a scatter plot of ozone residuals and temperature residuals indicates no relationship between them (Figure 3). Therefore, we proceed with our analysis of noise-free (filtered) temperature and ozone data. A linear regression of the filtered log of ozone concentrations [Okz(0] on filtered temperature [T^t)] data was performed:

and e(t) reveal a change in ozone concentrations in 1989 which cannot be attributed to temperature fluctuations (Figure 6). It should be noted that the ordinate in Figure 6 corresponds approximately to percent changes from the long-term mean. Thus, this QQ Plot of De-seasonalized Daily Maximum Temperature

(4)

R2 = 0.83 where a and b are fitted parameters, e(t) are the residuals of the relationship, and R2 is the square of the correlation coefficient. The averaged annual profiles of temperature and ozone, presented in Figure 4, indicate a phase lag between temperature and ozone. Ozone concentrations are influenced by both emissions and meteorological variables, whereas temperature is dictated primarily by the prevailing meteorological conditions. The linear relationship between O^t) and Tte(t) becomes stronger when the temperature data are lagged by 19 days (Figure 5).

QQ Plot of De-seasonalized Log of Ozone

(5)

R2 = 0.93 The e(t) reveal changes in ozone attributable to changes in emissions. Moreover, the relationship is devoid of the white noise that would characterize a regression approach using the raw data. A KZj year 3 filter was applied to e(t),

Ounmt d Smhnl Nomm

QQ Plot of 3,000 Normal Random Numbers

Figure 1. Seasonal variation in the daily maxima of temperature and log of ozone derived from the application of KZ2g3 to the original time series.

Figure 2. (a) Quantile-quantile plot of de-seasonalized daily maxima of temperature, (b) Same as (a), except for the log of ozone dally maxima, (c) Same as (a), except for 3,000 normal random numbers.

1090 • September 1994 • Vol. 44 • Journal of Air & Waste Management Association

Scatter Plot and Least-Squares Regressions! Pit

9Intercept 1.73439

slope 0.03974462

Rts eld dev 0.1265966

R-squared 0.930473

3-

M

Figure 3. Scatter diagram between de-seasonalized daily maxima of temperature and log of ozone.

70

Baamxrtl OaDy Mm Tanpmlura

Figure S. Linear least-squares regression fit for the seasonal components of temperature and log of ozone daily maxima. Log of Max Ozone with Temperature Effect Removed

Figure 4. Averaged annual profiles of temperature and ozone.

procedure reveals changes in ozone concentrations which may be attributable to changes in emissions. The original ozone time series, O(t), now may be defined as temperature-dependent time series, Oyjt), and the white noise process, W(t), (7)

Okz(t).

For the data set examined here, Wt and 0^(0 contribute approximately 20 and 80 percent, respectively, to the total variance in O(t). Substituting expressions (5) and (6) into expression (7), the ozone time series then is represented as Ot= • Wt+

19)

5(t).

(8)

In the right side of expression (8), the first term represents short-term turbulence which is always uncorrelated with longterm effects. The second term represents long-term and seasonal temperature effects in ozone, and the third term reflects long-term emissions effects unexplained by temperature. The fourth term, [8(t) = {e(t) - £KzlyMr,3(t)}], represents small ozone seasonal variations induced by'meteorological variables other than temperature. With the regression equation (5) explaining 93 percent of the variance in Oyjt), the contribution of the seasonal component, Tte(t +19), to the total variance in O(t) is approximately 70 percent (0.93 x 0.80).

Figure 6. The variation in log of ozone maxima when the temperature effect is removed from ozone concentrations [e(t) in expression (5)] along with the trend derived from the application of KZ, y M r 3 to e(t).

Expressions (4) and (7) are given for the natural logarithm of the original ozone data. An additive term e(t) then has a multiplicative effect, exp{e(t)}, in the original data. Since e(t) is sufficiently small, we have expe (t) * 1 + e(t); i.e., In (1 + e(t)) = e(t).

(9)

Hence, e(t)* 100 or 6(t)* 100 in Figure 6 corresponds to percent changes in long-term emissions effects in the original ozone data. The zero level of e(t) corresponds to the mean value of the total ozone data used for analyses. As a first step in our analysis, a logtransform of the original data was performed to stabilize the variance of noise W(t) in expression (7), which also has a multiplicative effect on the ozone concentrations (ppb).

Results and Discussion The presence of significant short-term variations and strong seasonality in temperature and ozone data make it difficult to draw

Journal of Air & Waste Management Association • Vol. 44 • September 1994 • 1091

TECHNICAL PAPER inferences using traditional statistical methods. For example, a plot of annual average ozone concentrations does not reveal a trend during the period 1983 through 1991. Similarly, a scatter plot of daily maxima of temperature versus ozone showed little relationship between these two variables. Our analysis of ozone data at Cliffside Park, New Jersey, reveals that the long-term and seasonal variations in the original time series account for approximately 10 and 70 percent, respectively, of the total variance in the original time series. The stochastic component (white noise) is approximately 20 percent of the total variance in the original time series data. Traditional statistical methods cannot perform well under conditions in which the effect that we want to discern is much smaller than that of the seasonal and stochastic components in the data. The primary advantage of the KZ filter is that it can separate short- and long-term variations in time series of meteorological and air quality data because of its very high frequency resolution characteristics. The strong relationship between KZ293(Tt+i9) and KZ29>3(Ot) indicates a significant linear component of the effect of temperature on ozone concentrations. This effect is absent from the time series of the residuals [e(t)] in expression (5) above. We have also applied this method to daily maxima of temperature and ozone for the ozone season only (i.e., April 15 to October 15), treating the data in the October 16 to April 14 period as missing data. As expected, analysis of data covering only the ozone season also has indicated a similar change in ozone levels during 1989. We also have applied this method to ozone data at another nearby monitoring location and found the results similar to those obtained at the Cliffside Park monitoring location. Since the influence of meteorology on ambient ozone levels has been moderated in this study, the change in ozone levels detected in 1989 might be attributable to changes in emissions due to regulatory actions, such as the fuel volatility control strategy implemented by the Northeast states in 1989.11 However, it is important to investigate whether other limiting factors, such as NOX availability, may have also contributed to the change in ozone levels in the post-1988 period. Furthermore, it is plausible that other meteorological variables, such as ventilation, water vapor, isolation, etc., may have also influenced ozone concentrations. Such effects are evidenced by the semi-annual cycle present in the temperature-independent ozone time series in Figure 6. If data are available, the influence of the above meteorological variables on ozone concentrations also may be removed similarly by this method. We currently are applying this technique to data spanning the period 1980 through 1992 from several monitoring stations, in an effort to demonstrate the method's robustness and to examine trends in ozone air quality in the northeastern United States. These results, when combined with trends in ozone precursor concentrations, should enable us to relate changes in ambient ozone air quality to changes in emissions and thus track progress toward ozone compliance. Summary In this paper, we presented a technique to detect changes in ozone air quality in the presence of meteorological fluctuations. The results indicate that the method can filter out or moderate the influence of meteorology on ozone, enabling us to track changes in ozone air quality due to changes in emissions. This then will allow us to evaluate the effectiveness of regulatory programs in improving ambient ozone air quality. Acknowledgments The authors gratefully acknowledge the support of Drs. Robert Eskridge and Thomas Karl of the Global Climate Laboratory, National Climatic Data Center, Asheville, North Carolina, for the development of the moving average filter used in this study, under

NOAA Contract 50EANE2000078. One of the authors (I.Z.) would like to acknowledge NIEHS Grant P42 ES04913-04 for partial support of the development of the filtering procedure. The authors also would like to thank professors Steven Porter of the University of Idaho and Peter Bloomfield of North Carolina State University for their critical review and constructive comments on the manuscript, and Mr. Shiqun Li for his assistance with the data analysis.

References 1. NRC, "Rethinking the ozone problem in urban and regional air pollution," National Academy Press (1991). 2. P.E. Korsog, G.T. Wolff, "An examination of urban ozone trends in the northeastern U.S. (1973-1983) using a robust method,"Atmos. Environ., 25B:47 (1991). 3. S.T. Rao, "Prepared discussion: Ozone air quality models," JAPCA, 38:1129(1988). 4. S.T. Rao, G. Sistla, R. Henry, "Statistical analysis of trends in urban ozone air quality," J. Air & Waste Manage. Assoc, 42:1204 (1992). 5. S.T. Rao, G. Sistla, K. Schere, J. Godowitch, "Analysis of ozone air quality over the New York metropolitan area," in Air Pollution Modeling and Its Applications VIII, H. van Dop, D.G. Steyn, eds., Plenum Press, NY (1991). 6. H.M. Walker, "Ten year trends in California and Texas," JAPCA, 35:903 (1985). 7. P.G. Wakim, "Temperature-adjusted ozone trends for Houston, New York, and Washington, D.C.," Paper No. 89-35.1, in Proc. 82ndAnnual APCA Meeting, Anaheim, CA (1989). 8. P. Bloomfield, A. Royle, Q. Yang, "Accounting for meteorological effects in measuring urban ozone levels and trends," Technical Report #1, National Institute of Statistical Sciences, P.O. Box 14162, Research Triangle Park, NC 27707, (June, 1993). 9. I. Zurbenko, The Spectral Analysis of Time Series, North Holland (1986). 10. I. Zurbenko, "Spectral analysis of nonstationary time series," International Statistical Review, 59:163, (1991). 11. NESCAUM, "Northeast states implement gasoline volatility regulations," Northeast Aireport, Issue No. 6 (1989).

About the Authors Dr. Rao is a research professor in the Department of Atmospheric Science at the State University of New York at Albany (SUNYA) and director of the Bureau of Air Research at the New York State Department of Environmental Conservation in Albany, NY 12233. Dr. Zurbenko is a professor of statistics in the Department of Biometry and Statistics of the School of Public Health at SUNYA, Albany, NY.

1092 • September 1994 • Vol. 44 • Journal of Air & Waste Management Association